summary-agent — Research Summarizer

Overview

The summary-agent is a lightweight, tool-free subagent that takes detailed research findings (typically from the draft-subagent) and condenses them into a clear, structured summary. This summary is included in the main agent's final response to give users a quick overview before the downloadable report.

Dictionary-Based SubAgent Definition:

python

summary_subagent = {
    "name": "summary-agent",
    "description": "Generates concise summary of the context from draftsubagent.",
    "system_prompt": "...",  # Full prompt below
    "model": subagent_model  # Default: gemini_3_flash
}

When the Main Agent Should Invoke This Subagent

•Tier 3 (Deep Research) — Step 5 (Summary), after the draft passes verification
•Called AFTER deep-reasoning-agent returns STATUS: VALID (or after max revisions)
•Called BEFORE report-subagent generates the final document

Invocation Pattern:

python

task(name="summary-agent", task="Summarize the following verified research draft on [topic]: [draft content]")

Tools

None. The summary-agent is a pure reasoning agent with no tool access.

Output Format

The summary follows a strict three-part structure:

markdown

## Summary of Findings
A comprehensive 3-paragraph summary that captures the essence of the research
findings without losing critical details.

Paragraph 1: [Overview of the main topic and scope of the research]

Paragraph 2: [Key discoveries, methodologies, and significant results]

Paragraph 3: [Implications, limitations, and areas for further investigation]

## Key Evidence
- Fact 1: [Specific finding with attribution]
- Fact 2: [Statistical evidence or data point]
- Fact 3: [Important comparison or contrast]
- Fact 4: [Technical insight or methodology finding]
- Fact 5: [Trend or pattern identified]
- Fact 6: [Notable limitation or gap]
- Fact 7: [Future direction or recommendation]

## Source List
- [Title 1](URL1)
- [Title 2](URL2)
- [Title 3](URL3)
- ...

Requirements

Attribute	Specification
Summary length	3 comprehensive paragraphs
Key Evidence	5–7 bullet points
Source List	Clean `[Title](URL)` format for every source
Tone	Clear, concise, professional
Coverage	Must capture essence without losing critical details

Pipeline Position

code

websearch-agent + academic-paper-agent
         ↓
    draft-subagent (drafting)
         ↓
    deep-reasoning-agent (verification)
         ↓
    summary-agent ← YOU ARE HERE
         ↓
    report-subagent (document generation)
         ↓
    Main agent final response (summary + download marker)

The summary output is included in the main agent's final response before the download marker, giving users a text overview of the research before they download the full report.

Integration with Final Response

The main agent's final response structure:

markdown

## Research Summary

[summary-agent output here — 3 paragraphs + key evidence + sources]

[DOWNLOAD_DOCX]{"filename": "...", "data": "..."} ← from report-subagent

Key Principles

•Concise but complete — Capture the essence, not the entirety
•Structured — Follow the exact three-part format
•Source-attributed — Every claim traced to a source
•Reader-friendly — Written for the end user, not for other agents
•No meta-commentary — Don't describe what you're doing, just do it